| 1 |
/* |
| 2 |
* Copyright (C) 2025 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2025 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 13 апр. 2025 г. |
| 7 |
* |
| 8 |
* lsp-dsp-lib is free software: you can redistribute it and/or modify |
| 9 |
* it under the terms of the GNU Lesser General Public License as published by |
| 10 |
* the Free Software Foundation, either version 3 of the License, or |
| 11 |
* any later version. |
| 12 |
* |
| 13 |
* lsp-dsp-lib is distributed in the hope that it will be useful, |
| 14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 |
* GNU Lesser General Public License for more details. |
| 17 |
* |
| 18 |
* You should have received a copy of the GNU Lesser General Public License |
| 19 |
* along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>. |
| 20 |
*/ |
| 21 |
|
| 22 |
#ifndef PRIVATE_DSP_ARCH_AARCH64_ASIMD_PMATH_SIN_H_ |
| 23 |
#define PRIVATE_DSP_ARCH_AARCH64_ASIMD_PMATH_SIN_H_ |
| 24 |
|
| 25 |
#ifndef PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL |
| 26 |
#error "This header should not be included directly" |
| 27 |
#endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL */ |
| 28 |
|
| 29 |
namespace lsp |
| 30 |
{ |
| 31 |
namespace asimd |
| 32 |
{ |
| 33 |
IF_ARCH_AARCH64( |
| 34 |
static const float kp_gen_const[] __lsp_aligned16 = |
| 35 |
{ |
| 36 |
0.0f, 1.0f, 2.0f, 3.0f, // +0x00: Initial values |
| 37 |
4.0f, 5.0f, 6.0f, 7.0f, |
| 38 |
LSP_DSP_VEC4(8.0), // +0x20: Step x8 |
| 39 |
LSP_DSP_VEC4(4.0), // +0x30: Step x4 |
| 40 |
}; |
| 41 |
|
| 42 |
static const uint32_t sinf_const[] __lsp_aligned16 = |
| 43 |
{ |
| 44 |
LSP_DSP_VEC4(0x3fc90fdb), // +0x000: PI/2 |
| 45 |
LSP_DSP_VEC4(0x40490fdb), // +0x010: PI |
| 46 |
LSP_DSP_VEC4(0x3e22f983), // +0x020: 1/(2*PI) |
| 47 |
LSP_DSP_VEC4(0xb2d7322b), // +0x030: C0 = -1/11! = -2.50521083854e-08 |
| 48 |
LSP_DSP_VEC4(0x3638ef1d), // +0x040: C1 = 1/9! = 2.7557319224e-06 |
| 49 |
LSP_DSP_VEC4(0xb9500d01), // +0x050: C2 = -1/7! = -0.000198412698413 |
| 50 |
LSP_DSP_VEC4(0x3c088889), // +0x060: C3 = 1/5! = 0.00833333333333 |
| 51 |
LSP_DSP_VEC4(0xbe2aaaab), // +0x070: C4 = -1/3! = -0.166666666667 |
| 52 |
}; |
| 53 |
) |
| 54 |
|
| 55 |
#define SINF_X_PLUS_PI_2_CORE_X16 \ |
| 56 |
/* v0 = X0 = x0 + PI/2 */ \ |
| 57 |
/* v1 = X1 = x1 + PI/2 */ \ |
| 58 |
/* v2 = X2 = x2 + PI/2 */ \ |
| 59 |
/* v3 = X3 = x3 + PI/2 */ \ |
| 60 |
/* v16 = PI/2 */ \ |
| 61 |
/* v17 = PI */ \ |
| 62 |
/* v18 = 1/(2*PI) */ \ |
| 63 |
/* v19 = C0 */ \ |
| 64 |
/* v20 = C1 */ \ |
| 65 |
/* v21 = C2 */ \ |
| 66 |
/* v22 = C3 */ \ |
| 67 |
/* v23 = C4 */ \ |
| 68 |
__ASM_EMIT("fmul v4.4s, v0.4s, v18.4s") /* v4 = x / (2*PI) */ \ |
| 69 |
__ASM_EMIT("fmul v5.4s, v1.4s, v18.4s") \ |
| 70 |
__ASM_EMIT("fmul v6.4s, v2.4s, v18.4s") \ |
| 71 |
__ASM_EMIT("fmul v7.4s, v3.4s, v18.4s") \ |
| 72 |
__ASM_EMIT("fcvtms v8.4s, v4.4s") /* v8 = int(X / (2*PI)) + [ X < 0.0 ] ? -1 : 0 */ \ |
| 73 |
__ASM_EMIT("fcvtms v9.4s, v5.4s") \ |
| 74 |
__ASM_EMIT("fcvtms v10.4s, v6.4s") \ |
| 75 |
__ASM_EMIT("fcvtms v11.4s, v7.4s") \ |
| 76 |
__ASM_EMIT("fadd v12.4s, v17.4s, v17.4s") /* v12 = 2*PI */ \ |
| 77 |
__ASM_EMIT("scvtf v8.4s, v8.4s") /* v8 = period = int(X / (2*PI)) [ X < 0.0 ] ? -1 : 0 */ \ |
| 78 |
__ASM_EMIT("scvtf v9.4s, v9.4s") \ |
| 79 |
__ASM_EMIT("scvtf v10.4s, v10.4s") \ |
| 80 |
__ASM_EMIT("scvtf v11.4s, v11.4s") \ |
| 81 |
__ASM_EMIT("fmls v0.4s, v8.4s, v12.4s") /* v0 = Y = X - period * 2 * PI */ \ |
| 82 |
__ASM_EMIT("fmls v1.4s, v9.4s, v12.4s") \ |
| 83 |
__ASM_EMIT("fmls v2.4s, v10.4s, v12.4s") \ |
| 84 |
__ASM_EMIT("fmls v3.4s, v11.4s, v12.4s") \ |
| 85 |
__ASM_EMIT("fadd v12.4s, v16.4s, v17.4s") /* v12 = 3*PI/2 */ \ |
| 86 |
__ASM_EMIT("fcmge v4.4s, v0.4s, v17.4s") /* v4 = [ Y >= PI ] */ \ |
| 87 |
__ASM_EMIT("fcmge v5.4s, v1.4s, v17.4s") \ |
| 88 |
__ASM_EMIT("fcmge v6.4s, v2.4s, v17.4s") \ |
| 89 |
__ASM_EMIT("fcmge v7.4s, v3.4s, v17.4s") \ |
| 90 |
__ASM_EMIT("fsub v8.4s, v12.4s, v0.4s") /* v8 = 3*PI/2 - Y */ \ |
| 91 |
__ASM_EMIT("fsub v9.4s, v12.4s, v1.4s") \ |
| 92 |
__ASM_EMIT("fsub v10.4s, v12.4s, v2.4s") \ |
| 93 |
__ASM_EMIT("fsub v11.4s, v12.4s, v3.4s") \ |
| 94 |
__ASM_EMIT("fsub v0.4s, v0.4s, v16.4s") /* v0 = Y - PI/2 */ \ |
| 95 |
__ASM_EMIT("fsub v1.4s, v1.4s, v16.4s") \ |
| 96 |
__ASM_EMIT("fsub v2.4s, v2.4s, v16.4s") \ |
| 97 |
__ASM_EMIT("fsub v3.4s, v3.4s, v16.4s") \ |
| 98 |
__ASM_EMIT("bit v0.16b, v8.16b, v4.16b") /* v0 = XX = [ Y < PI ] ? (Y - PI/2) : (3*PI/2 - Y) */ \ |
| 99 |
__ASM_EMIT("bit v1.16b, v9.16b, v5.16b") \ |
| 100 |
__ASM_EMIT("bit v2.16b, v10.16b, v6.16b") \ |
| 101 |
__ASM_EMIT("bit v3.16b, v11.16b, v7.16b") \ |
| 102 |
/* v0 = XX */ \ |
| 103 |
__ASM_EMIT("fmul v4.4s, v0.4s, v0.4s") /* v4 = X2 = XX*XX */ \ |
| 104 |
__ASM_EMIT("fmul v5.4s, v1.4s, v1.4s") \ |
| 105 |
__ASM_EMIT("fmul v6.4s, v2.4s, v2.4s") \ |
| 106 |
__ASM_EMIT("fmul v7.4s, v3.4s, v3.4s") \ |
| 107 |
__ASM_EMIT("fmul v8.4s, v4.4s, v19.4s") /* v8 = X2*C0 */ \ |
| 108 |
__ASM_EMIT("fmul v9.4s, v5.4s, v19.4s") \ |
| 109 |
__ASM_EMIT("fmul v10.4s, v6.4s, v19.4s") \ |
| 110 |
__ASM_EMIT("fmul v11.4s, v7.4s, v19.4s") \ |
| 111 |
__ASM_EMIT("fadd v8.4s, v8.4s, v20.4s") /* v8 = C1 + X2*C0 */ \ |
| 112 |
__ASM_EMIT("fadd v9.4s, v9.4s, v20.4s") \ |
| 113 |
__ASM_EMIT("fadd v10.4s, v10.4s, v20.4s") \ |
| 114 |
__ASM_EMIT("fadd v11.4s, v11.4s, v20.4s") \ |
| 115 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C1 + X2*C0) */ \ |
| 116 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 117 |
__ASM_EMIT("fmul v10.4s, v6.4s, v10.4s") \ |
| 118 |
__ASM_EMIT("fmul v11.4s, v7.4s, v11.4s") \ |
| 119 |
__ASM_EMIT("fadd v8.4s, v8.4s, v21.4s") /* v8 = C2 + X2*(C1 + X2*C0) */ \ |
| 120 |
__ASM_EMIT("fadd v9.4s, v9.4s, v21.4s") \ |
| 121 |
__ASM_EMIT("fadd v10.4s, v10.4s, v21.4s") \ |
| 122 |
__ASM_EMIT("fadd v11.4s, v11.4s, v21.4s") \ |
| 123 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 124 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 125 |
__ASM_EMIT("fmul v10.4s, v6.4s, v10.4s") \ |
| 126 |
__ASM_EMIT("fmul v11.4s, v7.4s, v11.4s") \ |
| 127 |
__ASM_EMIT("fadd v8.4s, v8.4s, v22.4s") /* v8 = C3 + X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 128 |
__ASM_EMIT("fadd v9.4s, v9.4s, v22.4s") \ |
| 129 |
__ASM_EMIT("fadd v10.4s, v10.4s, v22.4s") \ |
| 130 |
__ASM_EMIT("fadd v11.4s, v11.4s, v22.4s") \ |
| 131 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 132 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 133 |
__ASM_EMIT("fmul v10.4s, v6.4s, v10.4s") \ |
| 134 |
__ASM_EMIT("fmul v11.4s, v7.4s, v11.4s") \ |
| 135 |
__ASM_EMIT("fadd v8.4s, v8.4s, v23.4s") /* v8 = C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 136 |
__ASM_EMIT("fadd v9.4s, v9.4s, v23.4s") \ |
| 137 |
__ASM_EMIT("fadd v10.4s, v10.4s, v23.4s") \ |
| 138 |
__ASM_EMIT("fadd v11.4s, v11.4s, v23.4s") \ |
| 139 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 140 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 141 |
__ASM_EMIT("fmul v10.4s, v6.4s, v10.4s") \ |
| 142 |
__ASM_EMIT("fmul v11.4s, v7.4s, v11.4s") \ |
| 143 |
__ASM_EMIT("fmla v0.4s, v0.4s, v8.4s") /* v0 = XX*(1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))))) */ \ |
| 144 |
__ASM_EMIT("fmla v1.4s, v1.4s, v9.4s") \ |
| 145 |
__ASM_EMIT("fmla v2.4s, v2.4s, v10.4s") \ |
| 146 |
__ASM_EMIT("fmla v3.4s, v3.4s, v11.4s") |
| 147 |
|
| 148 |
#define SINF_X_PLUS_PI_2_CORE_X8 \ |
| 149 |
/* v0 = X0 = x0 + PI/2 */ \ |
| 150 |
/* v1 = X1 = x1 + PI/2 */ \ |
| 151 |
/* v16 = PI/2 */ \ |
| 152 |
/* v17 = PI */ \ |
| 153 |
/* v18 = 1/(2*PI) */ \ |
| 154 |
/* v19 = C0 */ \ |
| 155 |
/* v20 = C1 */ \ |
| 156 |
/* v21 = C2 */ \ |
| 157 |
/* v22 = C3 */ \ |
| 158 |
/* v23 = C4 */ \ |
| 159 |
__ASM_EMIT("fmul v4.4s, v0.4s, v18.4s") /* v4 = x / (2*PI) */ \ |
| 160 |
__ASM_EMIT("fmul v5.4s, v1.4s, v18.4s") \ |
| 161 |
__ASM_EMIT("fcvtms v8.4s, v4.4s") /* v8 = int(X / (2*PI)) + [ X < 0.0 ] ? -1 : 0 */ \ |
| 162 |
__ASM_EMIT("fcvtms v9.4s, v5.4s") \ |
| 163 |
__ASM_EMIT("fadd v12.4s, v17.4s, v17.4s") /* v12 = 2*PI */ \ |
| 164 |
__ASM_EMIT("scvtf v8.4s, v8.4s") /* v8 = period = int(X / (2*PI)) [ X < 0.0 ] ? -1 : 0 */ \ |
| 165 |
__ASM_EMIT("scvtf v9.4s, v9.4s") \ |
| 166 |
__ASM_EMIT("fmls v0.4s, v8.4s, v12.4s") /* v0 = Y = X - period * 2 * PI */ \ |
| 167 |
__ASM_EMIT("fmls v1.4s, v9.4s, v12.4s") \ |
| 168 |
__ASM_EMIT("fadd v12.4s, v16.4s, v17.4s") /* v12 = 3*PI/2 */ \ |
| 169 |
__ASM_EMIT("fcmge v4.4s, v0.4s, v17.4s") /* v4 = [ Y >= PI ] */ \ |
| 170 |
__ASM_EMIT("fcmge v5.4s, v1.4s, v17.4s") \ |
| 171 |
__ASM_EMIT("fsub v8.4s, v12.4s, v0.4s") /* v8 = 3*PI/2 - Y */ \ |
| 172 |
__ASM_EMIT("fsub v9.4s, v12.4s, v1.4s") \ |
| 173 |
__ASM_EMIT("fsub v0.4s, v0.4s, v16.4s") /* v0 = Y - PI/2 */ \ |
| 174 |
__ASM_EMIT("fsub v1.4s, v1.4s, v16.4s") \ |
| 175 |
__ASM_EMIT("bit v0.16b, v8.16b, v4.16b") /* v0 = XX = [ Y < PI ] ? (Y - PI/2) : (3*PI/2 - Y) */ \ |
| 176 |
__ASM_EMIT("bit v1.16b, v9.16b, v5.16b") \ |
| 177 |
/* v0 = XX */ \ |
| 178 |
__ASM_EMIT("fmul v4.4s, v0.4s, v0.4s") /* v4 = X2 = XX*XX */ \ |
| 179 |
__ASM_EMIT("fmul v5.4s, v1.4s, v1.4s") \ |
| 180 |
__ASM_EMIT("fmul v8.4s, v4.4s, v19.4s") /* v8 = X2*C0 */ \ |
| 181 |
__ASM_EMIT("fmul v9.4s, v5.4s, v19.4s") \ |
| 182 |
__ASM_EMIT("fadd v8.4s, v8.4s, v20.4s") /* v8 = C1 + X2*C0 */ \ |
| 183 |
__ASM_EMIT("fadd v9.4s, v9.4s, v20.4s") \ |
| 184 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C1 + X2*C0) */ \ |
| 185 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 186 |
__ASM_EMIT("fadd v8.4s, v8.4s, v21.4s") /* v8 = C2 + X2*(C1 + X2*C0) */ \ |
| 187 |
__ASM_EMIT("fadd v9.4s, v9.4s, v21.4s") \ |
| 188 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 189 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 190 |
__ASM_EMIT("fadd v8.4s, v8.4s, v22.4s") /* v8 = C3 + X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 191 |
__ASM_EMIT("fadd v9.4s, v9.4s, v22.4s") \ |
| 192 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 193 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 194 |
__ASM_EMIT("fadd v8.4s, v8.4s, v23.4s") /* v8 = C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 195 |
__ASM_EMIT("fadd v9.4s, v9.4s, v23.4s") \ |
| 196 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 197 |
__ASM_EMIT("fmul v9.4s, v5.4s, v9.4s") \ |
| 198 |
__ASM_EMIT("fmla v0.4s, v0.4s, v8.4s") /* v0 = XX*(1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))))) */ \ |
| 199 |
__ASM_EMIT("fmla v1.4s, v1.4s, v9.4s") |
| 200 |
|
| 201 |
#define SINF_X_PLUS_PI_2_CORE_X4 \ |
| 202 |
/* v0 = X = x + PI/2 */ \ |
| 203 |
/* v16 = PI/2 */ \ |
| 204 |
/* v17 = PI */ \ |
| 205 |
/* v18 = 1/(2*PI) */ \ |
| 206 |
/* v19 = C0 */ \ |
| 207 |
/* v20 = C1 */ \ |
| 208 |
/* v21 = C2 */ \ |
| 209 |
/* v22 = C3 */ \ |
| 210 |
/* v23 = C4 */ \ |
| 211 |
__ASM_EMIT("fmul v4.4s, v0.4s, v18.4s") /* v4 = x / (2*PI) */ \ |
| 212 |
__ASM_EMIT("fcvtms v8.4s, v4.4s") /* v8 = int(X / (2*PI)) + [ X < 0.0 ] ? -1 : 0 */ \ |
| 213 |
__ASM_EMIT("fadd v12.4s, v17.4s, v17.4s") /* v12 = 2*PI */ \ |
| 214 |
__ASM_EMIT("scvtf v8.4s, v8.4s") /* v8 = period = int(X / (2*PI)) [ X < 0.0 ] ? -1 : 0 */ \ |
| 215 |
__ASM_EMIT("fmls v0.4s, v8.4s, v12.4s") /* v0 = Y = X - period * 2 * PI */ \ |
| 216 |
__ASM_EMIT("fadd v12.4s, v16.4s, v17.4s") /* v12 = 3*PI/2 */ \ |
| 217 |
__ASM_EMIT("fcmge v4.4s, v0.4s, v17.4s") /* v4 = [ Y >= PI ] */ \ |
| 218 |
__ASM_EMIT("fsub v8.4s, v12.4s, v0.4s") /* v8 = 3*PI/2 - Y */ \ |
| 219 |
__ASM_EMIT("fsub v0.4s, v0.4s, v16.4s") /* v0 = Y - PI/2 */ \ |
| 220 |
__ASM_EMIT("bit v0.16b, v8.16b, v4.16b") /* v0 = XX = [ Y < PI ] ? (Y - PI/2) : (3*PI/2 - Y) */ \ |
| 221 |
/* v0 = XX */ \ |
| 222 |
__ASM_EMIT("fmul v4.4s, v0.4s, v0.4s") /* v4 = X2 = XX*XX */ \ |
| 223 |
__ASM_EMIT("fmul v8.4s, v4.4s, v19.4s") /* v8 = X2*C0 */ \ |
| 224 |
__ASM_EMIT("fadd v8.4s, v8.4s, v20.4s") /* v8 = C1 + X2*C0 */ \ |
| 225 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C1 + X2*C0) */ \ |
| 226 |
__ASM_EMIT("fadd v8.4s, v8.4s, v21.4s") /* v8 = C2 + X2*(C1 + X2*C0) */ \ |
| 227 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 228 |
__ASM_EMIT("fadd v8.4s, v8.4s, v22.4s") /* v8 = C3 + X2*(C2 + X2*(C1 + X2*C0)) */ \ |
| 229 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 230 |
__ASM_EMIT("fadd v8.4s, v8.4s, v23.4s") /* v8 = C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))) */ \ |
| 231 |
__ASM_EMIT("fmul v8.4s, v4.4s, v8.4s") /* v8 = X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0)))) */ \ |
| 232 |
__ASM_EMIT("fmla v0.4s, v0.4s, v8.4s") /* v0 = XX*(1.0 + X2*(C4 + X2*(C3 + X2*(C2 + X2*(C1 + X2*C0))))) */ |
| 233 |
|
| 234 |
void sinf1(float *dst, size_t count) |
| 235 |
{ |
| 236 |
ARCH_AARCH64_ASM( |
| 237 |
__ASM_EMIT("ldp q16, q17, [%[S2C], #0x00]") /* v16 = PI/2, v17 = PI */ |
| 238 |
__ASM_EMIT("ldp q18, q19, [%[S2C], #0x20]") /* v18 = 1/(2*PI), v19 = C0 */ |
| 239 |
__ASM_EMIT("ldp q20, q21, [%[S2C], #0x40]") /* v20 = C1, v21 = C2 */ |
| 240 |
__ASM_EMIT("ldp q22, q23, [%[S2C], #0x60]") /* v22 = C3, v23 = C4 */ |
| 241 |
// x16 blocks |
| 242 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 243 |
__ASM_EMIT("b.lo 2f") |
| 244 |
__ASM_EMIT("1:") |
| 245 |
__ASM_EMIT("ldp q0, q1, [%[dst], #0x00]") // v0 = x0, v1 = x1 |
| 246 |
__ASM_EMIT("ldp q2, q3, [%[dst], #0x20]") // v2 = x2, v3 = x3 |
| 247 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 248 |
__ASM_EMIT("fadd v1.4s, v1.4s, v16.4s") // v1 = x1 + PI/2 |
| 249 |
__ASM_EMIT("fadd v2.4s, v2.4s, v16.4s") // v2 = x2 + PI/2 |
| 250 |
__ASM_EMIT("fadd v3.4s, v3.4s, v16.4s") // v3 = x3 + PI/2 |
| 251 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 252 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 253 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 254 |
__ASM_EMIT("stp q2, q3, [%[dst], #0x20]") |
| 255 |
__ASM_EMIT("add %[dst], %[dst], #0x40") |
| 256 |
__ASM_EMIT("b.hs 1b") |
| 257 |
__ASM_EMIT("2:") |
| 258 |
// x8 block |
| 259 |
__ASM_EMIT("adds %[count], %[count], #8") |
| 260 |
__ASM_EMIT("b.lt 4f") |
| 261 |
__ASM_EMIT("ldp q0, q1, [%[dst], #0x00]") // v0 = x0, v1 = x1 |
| 262 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 263 |
__ASM_EMIT("fadd v1.4s, v1.4s, v16.4s") // v1 = x1 + PI/2 |
| 264 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 265 |
__ASM_EMIT("sub %[count], %[count], #8") |
| 266 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 267 |
__ASM_EMIT("add %[dst], %[dst], #0x20") |
| 268 |
__ASM_EMIT("4:") |
| 269 |
// x4 block |
| 270 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 271 |
__ASM_EMIT("b.lt 6f") |
| 272 |
__ASM_EMIT("ldr q0, [%[dst], #0x00]") // v0 = x0 |
| 273 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 274 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 275 |
__ASM_EMIT("sub %[count], %[count], #4") |
| 276 |
__ASM_EMIT("str q0, [%[dst], #0x00]") |
| 277 |
__ASM_EMIT("add %[dst], %[dst], #0x10") |
| 278 |
__ASM_EMIT("6:") |
| 279 |
// Tail: 1x-3x block |
| 280 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 281 |
__ASM_EMIT("b.ls 14f") |
| 282 |
__ASM_EMIT("tst %[count], #1") |
| 283 |
__ASM_EMIT("b.eq 8f") |
| 284 |
__ASM_EMIT("ld1 {v0.s}[0], [%[dst]]") |
| 285 |
__ASM_EMIT("add %[dst], %[dst], #0x04") |
| 286 |
__ASM_EMIT("8:") |
| 287 |
__ASM_EMIT("tst %[count], #2") |
| 288 |
__ASM_EMIT("b.eq 10f") |
| 289 |
__ASM_EMIT("ld1 {v0.d}[1], [%[dst]]") |
| 290 |
__ASM_EMIT("10:") |
| 291 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 292 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 293 |
__ASM_EMIT("tst %[count], #1") |
| 294 |
__ASM_EMIT("b.eq 12f") |
| 295 |
__ASM_EMIT("sub %[dst], %[dst], #0x04") |
| 296 |
__ASM_EMIT("st1 {v0.s}[0], [%[dst]]") |
| 297 |
__ASM_EMIT("add %[dst], %[dst], #0x04") |
| 298 |
__ASM_EMIT("12:") |
| 299 |
__ASM_EMIT("tst %[count], #2") |
| 300 |
__ASM_EMIT("b.eq 14f") |
| 301 |
__ASM_EMIT("st1 {v0.d}[1], [%[dst]]") |
| 302 |
// End |
| 303 |
__ASM_EMIT("14:") |
| 304 |
|
| 305 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 306 |
: [S2C] "r" (&sinf_const[0]) |
| 307 |
: "cc", "memory", |
| 308 |
"v0", "v1", "v2", "v3", |
| 309 |
"v4", "v5", "v6", "v7", |
| 310 |
"v8", "v9", "v10", "v11", |
| 311 |
"v12", "v13", "v14", "v15", |
| 312 |
"v16", "v17", "v18", "v19", |
| 313 |
"v20", "v21", "v22", "v23" |
| 314 |
); |
| 315 |
} |
| 316 |
|
| 317 |
void sinf2(float *dst, const float *src, size_t count) |
| 318 |
{ |
| 319 |
ARCH_AARCH64_ASM( |
| 320 |
__ASM_EMIT("ldp q16, q17, [%[S2C], #0x00]") /* v16 = PI/2, v17 = PI */ |
| 321 |
__ASM_EMIT("ldp q18, q19, [%[S2C], #0x20]") /* v18 = 1/(2*PI), v19 = C0 */ |
| 322 |
__ASM_EMIT("ldp q20, q21, [%[S2C], #0x40]") /* v20 = C1, v21 = C2 */ |
| 323 |
__ASM_EMIT("ldp q22, q23, [%[S2C], #0x60]") /* v22 = C3, v23 = C4 */ |
| 324 |
// x16 blocks |
| 325 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 326 |
__ASM_EMIT("b.lo 2f") |
| 327 |
__ASM_EMIT("1:") |
| 328 |
__ASM_EMIT("ldp q0, q1, [%[src], #0x00]") // v0 = x0, v1 = x1 |
| 329 |
__ASM_EMIT("ldp q2, q3, [%[src], #0x20]") // v2 = x2, v3 = x3 |
| 330 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 331 |
__ASM_EMIT("fadd v1.4s, v1.4s, v16.4s") // v1 = x1 + PI/2 |
| 332 |
__ASM_EMIT("fadd v2.4s, v2.4s, v16.4s") // v2 = x2 + PI/2 |
| 333 |
__ASM_EMIT("fadd v3.4s, v3.4s, v16.4s") // v3 = x3 + PI/2 |
| 334 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 335 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 336 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 337 |
__ASM_EMIT("stp q2, q3, [%[dst], #0x20]") |
| 338 |
__ASM_EMIT("add %[src], %[src], #0x40") |
| 339 |
__ASM_EMIT("add %[dst], %[dst], #0x40") |
| 340 |
__ASM_EMIT("b.hs 1b") |
| 341 |
__ASM_EMIT("2:") |
| 342 |
// x8 block |
| 343 |
__ASM_EMIT("adds %[count], %[count], #8") |
| 344 |
__ASM_EMIT("b.lt 4f") |
| 345 |
__ASM_EMIT("ldp q0, q1, [%[src], #0x00]") // v0 = x0, v1 = x1 |
| 346 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 347 |
__ASM_EMIT("fadd v1.4s, v1.4s, v16.4s") // v1 = x1 + PI/2 |
| 348 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 349 |
__ASM_EMIT("sub %[count], %[count], #8") |
| 350 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 351 |
__ASM_EMIT("add %[src], %[src], #0x20") |
| 352 |
__ASM_EMIT("add %[dst], %[dst], #0x20") |
| 353 |
__ASM_EMIT("4:") |
| 354 |
// x4 block |
| 355 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 356 |
__ASM_EMIT("b.lt 6f") |
| 357 |
__ASM_EMIT("ldr q0, [%[src], #0x00]") // v0 = x0 |
| 358 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 359 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 360 |
__ASM_EMIT("sub %[count], %[count], #4") |
| 361 |
__ASM_EMIT("str q0, [%[dst], #0x00]") |
| 362 |
__ASM_EMIT("add %[src], %[src], #0x10") |
| 363 |
__ASM_EMIT("add %[dst], %[dst], #0x10") |
| 364 |
__ASM_EMIT("6:") |
| 365 |
// Tail: 1x-3x block |
| 366 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 367 |
__ASM_EMIT("b.ls 14f") |
| 368 |
__ASM_EMIT("tst %[count], #1") |
| 369 |
__ASM_EMIT("b.eq 8f") |
| 370 |
__ASM_EMIT("ld1 {v0.s}[0], [%[src]]") |
| 371 |
__ASM_EMIT("add %[src], %[src], #0x04") |
| 372 |
__ASM_EMIT("8:") |
| 373 |
__ASM_EMIT("tst %[count], #2") |
| 374 |
__ASM_EMIT("b.eq 10f") |
| 375 |
__ASM_EMIT("ld1 {v0.d}[1], [%[src]]") |
| 376 |
__ASM_EMIT("10:") |
| 377 |
__ASM_EMIT("fadd v0.4s, v0.4s, v16.4s") // v0 = x0 + PI/2 |
| 378 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 379 |
__ASM_EMIT("tst %[count], #1") |
| 380 |
__ASM_EMIT("b.eq 12f") |
| 381 |
__ASM_EMIT("st1 {v0.s}[0], [%[dst]]") |
| 382 |
__ASM_EMIT("add %[dst], %[dst], #0x04") |
| 383 |
__ASM_EMIT("12:") |
| 384 |
__ASM_EMIT("tst %[count], #2") |
| 385 |
__ASM_EMIT("b.eq 14f") |
| 386 |
__ASM_EMIT("st1 {v0.d}[1], [%[dst]]") |
| 387 |
// End |
| 388 |
__ASM_EMIT("14:") |
| 389 |
|
| 390 |
: [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) |
| 391 |
: [S2C] "r" (&sinf_const[0]) |
| 392 |
: "cc", "memory", |
| 393 |
"v0", "v1", "v2", "v3", |
| 394 |
"v4", "v5", "v6", "v7", |
| 395 |
"v8", "v9", "v10", "v11", |
| 396 |
"v12", "v13", "v14", "v15", |
| 397 |
"v16", "v17", "v18", "v19", |
| 398 |
"v20", "v21", "v22", "v23" |
| 399 |
); |
| 400 |
} |
| 401 |
|
| 402 |
void sinf_kp1(float *dst, float k, float p, size_t count) |
| 403 |
{ |
| 404 |
ARCH_AARCH64_ASM( |
| 405 |
__ASM_EMIT("ldp q16, q17, [%[S2C], #0x00]") /* v16 = PI/2, v17 = PI */ |
| 406 |
__ASM_EMIT("ldp q18, q19, [%[S2C], #0x20]") /* v18 = 1/(2*PI), v19 = C0 */ |
| 407 |
__ASM_EMIT("ldp q20, q21, [%[S2C], #0x40]") /* v20 = C1, v21 = C2 */ |
| 408 |
__ASM_EMIT("ldp q22, q23, [%[S2C], #0x60]") /* v22 = C3, v23 = C4 */ |
| 409 |
__ASM_EMIT("ld1r {v28.4s}, [%[k]]") /* v28 = k k k k */ |
| 410 |
__ASM_EMIT("ld1r {v29.4s}, [%[p]]") /* v29 = p p p p */ |
| 411 |
__ASM_EMIT("ldp q24, q25, [%[S2KP], #0x00]") /* v24 = 0 1 2 3, v25 = 4 5 6 7 */ |
| 412 |
__ASM_EMIT("ldp q26, q27, [%[S2KP], #0x20]") /* v26 = 8 8 8 8, v27 = 4 4 4 4 */ |
| 413 |
__ASM_EMIT("fadd v29.4s, v29.4s, v16.4s") /* v29 = p + PI/2 ... */ |
| 414 |
// x16 blocks |
| 415 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 416 |
__ASM_EMIT("b.lo 2f") |
| 417 |
__ASM_EMIT("1:") |
| 418 |
__ASM_EMIT("fmul v0.4s, v28.4s, v24.4s") // v0 = k*x0 |
| 419 |
__ASM_EMIT("fmul v1.4s, v28.4s, v25.4s") // v1 = k*x1 |
| 420 |
__ASM_EMIT("fadd v24.4s, v24.4s, v26.4s") // v24 = i0 + 8 |
| 421 |
__ASM_EMIT("fadd v25.4s, v25.4s, v26.4s") // v25 = i1 + 8 |
| 422 |
__ASM_EMIT("fadd v0.4s, v0.4s, v29.4s") // v0 = k*x0 + p + PI/2 |
| 423 |
__ASM_EMIT("fadd v1.4s, v1.4s, v29.4s") // v1 = k*x1 + p + PI/2 |
| 424 |
__ASM_EMIT("fmul v2.4s, v28.4s, v24.4s") // v2 = k*x2 |
| 425 |
__ASM_EMIT("fmul v3.4s, v28.4s, v25.4s") // v3 = k*x3 |
| 426 |
__ASM_EMIT("fadd v24.4s, v24.4s, v26.4s") // v24 = i0 + 8 |
| 427 |
__ASM_EMIT("fadd v25.4s, v25.4s, v26.4s") // v25 = i1 + 8 |
| 428 |
__ASM_EMIT("fadd v2.4s, v2.4s, v29.4s") // v2 = k*x2 + p + PI/2 |
| 429 |
__ASM_EMIT("fadd v3.4s, v3.4s, v29.4s") // v3 = k*x3 + p + PI/2 |
| 430 |
SINF_X_PLUS_PI_2_CORE_X16 |
| 431 |
__ASM_EMIT("subs %[count], %[count], #16") |
| 432 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 433 |
__ASM_EMIT("stp q2, q3, [%[dst], #0x20]") |
| 434 |
__ASM_EMIT("add %[dst], %[dst], #0x40") |
| 435 |
__ASM_EMIT("b.hs 1b") |
| 436 |
__ASM_EMIT("2:") |
| 437 |
// x8 block |
| 438 |
__ASM_EMIT("adds %[count], %[count], #8") |
| 439 |
__ASM_EMIT("b.lt 4f") |
| 440 |
__ASM_EMIT("fmul v0.4s, v28.4s, v24.4s") // v0 = k*x0 |
| 441 |
__ASM_EMIT("fmul v1.4s, v28.4s, v25.4s") // v1 = k*x1 |
| 442 |
__ASM_EMIT("fadd v24.4s, v24.4s, v26.4s") // v24 = i0 + 8 |
| 443 |
__ASM_EMIT("fadd v25.4s, v25.4s, v26.4s") // v25 = i1 + 8 |
| 444 |
__ASM_EMIT("fadd v0.4s, v0.4s, v29.4s") // v0 = k*x0 + p + PI/2 |
| 445 |
__ASM_EMIT("fadd v1.4s, v1.4s, v29.4s") // v1 = k*x1 + p + PI/2 |
| 446 |
SINF_X_PLUS_PI_2_CORE_X8 |
| 447 |
__ASM_EMIT("sub %[count], %[count], #8") |
| 448 |
__ASM_EMIT("stp q0, q1, [%[dst], #0x00]") |
| 449 |
__ASM_EMIT("add %[dst], %[dst], #0x20") |
| 450 |
__ASM_EMIT("4:") |
| 451 |
// x4 block |
| 452 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 453 |
__ASM_EMIT("b.lt 6f") |
| 454 |
__ASM_EMIT("fmul v0.4s, v28.4s, v24.4s") // v0 = k*x0 |
| 455 |
__ASM_EMIT("fadd v24.4s, v24.4s, v27.4s") // v24 = i0 + 4 |
| 456 |
__ASM_EMIT("fadd v0.4s, v0.4s, v29.4s") // v0 = k*x0 + p + PI/2 |
| 457 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 458 |
__ASM_EMIT("sub %[count], %[count], #4") |
| 459 |
__ASM_EMIT("str q0, [%[dst], #0x00]") |
| 460 |
__ASM_EMIT("add %[dst], %[dst], #0x10") |
| 461 |
__ASM_EMIT("6:") |
| 462 |
// Tail: 1x-3x block |
| 463 |
__ASM_EMIT("adds %[count], %[count], #4") |
| 464 |
__ASM_EMIT("b.ls 10f") |
| 465 |
__ASM_EMIT("fmul v0.4s, v28.4s, v24.4s") // v0 = k*x0 |
| 466 |
__ASM_EMIT("fadd v0.4s, v0.4s, v29.4s") // v0 = k*x0 + p + PI/2 |
| 467 |
SINF_X_PLUS_PI_2_CORE_X4 |
| 468 |
__ASM_EMIT("tst %[count], #1") |
| 469 |
__ASM_EMIT("b.eq 8f") |
| 470 |
__ASM_EMIT("st1 {v0.s}[0], [%[dst]]") |
| 471 |
__ASM_EMIT("add %[dst], %[dst], #0x04") |
| 472 |
__ASM_EMIT("ext v0.16b, v0.16b, v0.16b, #4") // v0 = S1 S2 S3 S0 |
| 473 |
__ASM_EMIT("8:") |
| 474 |
__ASM_EMIT("tst %[count], #2") |
| 475 |
__ASM_EMIT("b.eq 10f") |
| 476 |
__ASM_EMIT("st1 {v0.d}[0], [%[dst]]") |
| 477 |
// End |
| 478 |
__ASM_EMIT("10:") |
| 479 |
|
| 480 |
: [dst] "+r" (dst), [count] "+r" (count) |
| 481 |
: [k] "r" (&k), |
| 482 |
[p] "r" (&p), |
| 483 |
[S2C] "r" (&sinf_const[0]), |
| 484 |
[S2KP] "r" (&kp_gen_const[0]) |
| 485 |
: "cc", "memory", |
| 486 |
"v0", "v1", "v2", "v3", |
| 487 |
"v4", "v5", "v6", "v7", |
| 488 |
"v8", "v9", "v10", "v11", |
| 489 |
"v12", "v13", "v14", "v15", |
| 490 |
"v16", "v17", "v18", "v19", |
| 491 |
"v20", "v21", "v22", "v23", |
| 492 |
"v24", "v25", "v26", "v27", |
| 493 |
"v28", "v29" |
| 494 |
); |
| 495 |
} |
| 496 |
|
| 497 |
} /* namespace asimd */ |
| 498 |
} /* namespace lsp */ |
| 499 |
|
| 500 |
|
| 501 |
|
| 502 |
#endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_PMATH_SIN_H_ */ |